Recommend API with vectors/embeddings (instead of document id)

[igoristic]

Would be awesome if we can use vectors/embeddings in the recommend API, eg:

POST /collections/{collection_name}/points/recommend

{
  "negative": [[0.34, 0.11, 0.4, -0.15, ...], [-0.2, -0.33, 0.4, 0.15, ...]],
  "positive": [[0.11, -0.19, 0.8, 0.9, ...], [ ... ]],
  "limit": 10
}

Right now the only way to achieve this is to first put those embeddings into Qdrant and then use their ids